With increasing demands for audio recognition and an audio analysis, a technology for accurately analyzing audio generated by a speaker is desired. A method of the technology of the audio analysis is binary masking. In the binary masking, a frequency analysis is performed for each piece of audio obtained by a plurality of input devices, an input of a desired sound having a large signal level and an input of an undesired sound having a small signal level (noise or the like other than the desired sound) are specified by comparing magnitude of a signal level for each of frequency components, and an analysis of the desired sound is performed by removing the undesired sound.
Japanese Laid-open Patent Publication No. 2009-20471 is an example of the related art.